446 T. Wu
real-time critical computations, but also give an extra reliability and accuracy for
general-purpose numerical computations. This result can be provided by interval
computation [11, 12] too. But using the primitive Ada programming method can
eliminate the more complicated implementation of interval computation.
34.7 Conclusions
In this chapter, we have reviewed some basic algebraic structures such as abelian
group, ring,andfield. We defined sets of rough numbers, dyadic numbers, finite
dyadic numbers, limited dyadic numbers,andmodel numbers; from these defini-
tions, we have learned that the difficulty of numerical computation is one must
actually work with two distinct number systems. Solving any numerical compu-
tation problem consists of the following three parts: (1) the problem is given in the
real number system, (2) the computation is done in the model number system for the
given machine, and (3) the results must be converted from model numbers into real
numbers. Real numbers and model numbers have two different algebraic structures,
and they are not isomorphic. In general, starting from the problem to computation
on a machine can create some errors and going from the machine computation to
reporting results in real numbers can induce errors too. This chapter has reported
that a computer system is not capable of computing real numbers accurately within
its constraints from a theoretical viewpoint.
Computation over the set of real numbers requires performing a field computa-
tion. However, when a real number within the given range is stored or read into a
computer system, it is converted into a dyadic number. Computation over the set of
limited dyadic numbers is dyadic number computation.
Let f be a mapping that takes each real number into its dyadic number repre-
sentation. If a and b are two real numbers within a given range and they map into
their dyadic representations f.a/and f.b/, respectively, then in general, we should
have f.a C b/ ¤ f.a/ ˚ f.b/. The addition, C, on the right-hand side is the ad-
dition of real numbers. The mapping f does not preserve the algebraic structure,
so the set of real numbers and the set of dyadic numbers are not isomorphic.To
avoid overloading and possible confusion, we will introduce a new addition ‘˚’for
the dyadic number addition. To adjust the inequality, we will add in an error term,
Err. Therefore, we have f.a C b/ D f.a/ ˚ f.b/C Err. To do this, we use the
special features of the Ada programming language either to declare variables with
a specific value of “small”forfixed-point numbers or to designate a value of delta
for float-point numbers. In this way, the error of the computation result is controlled
within the acceptable level.